Issue during test stage when load_from_checkpoint

Trainer
1

hi guys! I’ve already accomplished training stage like this:
model = PL_Model(cfg)
trainer = Trainer(…)
trainer.fit(model)
now I need to test some data on the trained model, so I add
def test_step() and def test_end()
into the class PL_Model(), and test like this:
model = PL_Model(cfg).load_from_checkpoint(PATH)
trainer = Trainer(…)
trainer.test(model)
the error is:
File “…/lib/python3.6/site-packages/torch/nn/modules/module.py”, line 585, in getattr
type(self).name, name))
AttributeError: ‘PL_Model’ object has no attribute ‘load_from_checkpoint’
how can i fix it?

2

mind reproduce it with Google Colab??

3

thanks for your reply!!
but are there any solutions to test separately with an existed .ckpt file without AttributeError above?

4

the above error seems strange and should not occur since load_from_checkpoint is a classmethod in LightningModule. So I’m not sure why it’s not working for you. That’s why I requested a minimal reproducable code to check the issue.

5

I meet the same problem when using jupyter. After restart the jupyter, the problem is gone. I donot know why. But maybe it deserve to try Just Restart OR Reinstall.

1 Like
6

Personally, I found this problem is occurring because of the way I imported the model.
Fixed by changing:

import models.MyNetwork as MyNetwork

To:

from models.MyNetwork import MyNetwork